Proceedings of the 4th Science Science &
Engineering Conference on Sports Innovation
Eindhoven, The Netherlands, October 11, 2019
Edited by
Steven Vos,
Juan Restrepo Villamizar, and
Aarnout Brombacher.
Crossing Borders in Research
on Sport and Physical Activity
Crossing Borders in Research
on Sports and Physical Activity
Proceedings of the 4
thScience & Engineering
Conference on Sports Innovation
Eindhoven, The Netherlands, October 11
th,2019
Organized by
Edited by
prof.dr. Steven Vos
prof.dr. Aarnout Bombache
M.Sc. Juan Restrepo-Villamizar
ISBN: 978-90-386-4947-4
ISBN: 978-90-386-4947-4
Table of Contents
Introduction
Improving and Optimizing Sports Performance
Fully Automated Dynamic Ultrasound Muscle Analysis during Walking Exercise
Cristina Caresio, Benjamin Tchang, Jori Verbeek, Frans van de Vosse and Richard LopataThe Running & Mental Break Optimization (REMBO) App: Design Process and User
Evaluation
Luuk P. van Iperen , Jan de Jonge, Josette M.P. Gevers, and Steven B. Vos
Sports, Data and Coaching
Comparing the Performance of Youth Footballers on Anticipatory Capability and
their Respective Coach Ratings
Tom de Joode, Saumya A. Mehta, John van der Kamp and Geert J. P. Savelsbergh
Inspirun app: User Test of an Algorithm that Automatically adjusts Training
Sessions for Runners
Mark Janssen, Jos Goudsmit, and Steven Vos
Automatic extraction of performance metrics from football players with data
mining
Josá Carlos Coutinho, Marck de Greeff, Cláudio Rebelo de Sá 3, Nicolette Schipper-van Veldhoven
Stimulating Physical Activity
Grace: Designing for Exercise Motivation Through Social Support and Graceful
Interactions
Daphne Menheere, Carine Lallemand, Ilse Faber, Jesse Paping, Bram Monkel, Stella Xu and Steven Vos
Virtual Community Building in Organized Sports in the Netherlands
Nanny Kuijsters- Timmers, John Goedee and Roger LeendersCOMMONS: facilitating interdisciplinary collaboration in developing wearable
technology for physical activity
Dennis Arts, Len Kromkamp1 and Steven Vos
Sensing
The Introduction and Concurrent Validation of an Inertial Magnetic
Measurement-based Motion Tracking System in Soccer
Erik Wilme, Jo de Ruiter , Jasper van Zon and Geert Savelsbergh
11
15
19
23
27
33
37
41
47
53
1
2
3
4
5
Smart Sensor Shorts: a Novel IMU Based Method to Continuously Assess the
Biomechanical Training and Match Load in Team Sports.
Bram Bastiaansen, Michel Brink, Riemer Vegter and Koen Lemmink
Feedback
Designing Motor Learning Based Instruction and Feedback for Running
Technique Changes
Jos Goudsmit and Steven Vos
Who is active at Work? Expressing Social Feedback on Physical Activity in the
Office Environment
Hans Brombacher, Dennis Arts and Steven Vos
Synchronizing Steps in Running: Pros and Cons
Harjo J. de Poel1, Eric van der Meer, Frank Blikslager and Niek Blikslager
Designing Digitally-Augmented Feedback for Physical Education
Yudan Ma, Bin Yu, Steven Vos, Joep van de Ven, Tilde Bekker and Jun HuPerception of Vibrotactile Feedback in Cycling: Development of an Indoor
Training Bike System for Correcting the Aerodynamic Position
Thomas Peeters, Jochen Vleugels, Steven Truijen and Stijn Verwulgen
Haptic Feedback in Running: Can we Use Muscle Stimulation for Information
Transfer?
Kevin Lu and Aarnout Brombacher
Towards Translating the Effect of Music into Motor Using Deep Learning
Olaf T.A. Janssen, Tom Langhorst and Bernd-Jan WitkampExploring the Value of User-generated App Data to Design and Improve Urban
Running Environments
Loes van Renswouw, Sander Bogers, Carine Lallemand and Steven Vos
Psycho-physical Implications of On-Skin Computing Interfaces for Sports and
Physical Activity
Juan Restrepo-Villamizar, Evert Verhagen and Steven Vos
61
67
71
75
79
83
87
91
95
99
6
Wind tunnel facilities
Eindhoven University of Technology
Dear Reader,
On Friday 11 October 2019, the 4th edition of the Science and Engineering Conference
on Sport Innovations (SECSI2019) took place in Eindhoven. The organization of
SECSI2019 was this year in the hands of Eindhoven University of Technology, faculty
Industrial Design, and Fontys University of Applied Sciences, School of Sport Studies.
With the exchange of knowledge, this conference aims to stimulate sports research and
innovation. Furthermore, its goal is to strengthen the partnerships between universities,
research institutions, and universities of applied science in the Low Countries.
The participants of SECSI2019 got state-of-the-art insights in no less than 3 keynotes and
21 presentations in 7 parallel sessions. During these presentations, research results from
disciplines such as human movement sciences, psychology, design, engineering, data
science, etcetera. were shared. Extended abstracts of these presentations can be found
in this proceedings.
Dr. Nicole Ummelen, vice president of the Executive Board of Eindhoven University of
Technology, welcomed the participants and opened SECSI2019. Subsequently, the first
keynote session entitled ‘Technology and data in football: the missing link’ was given
by Professor Koen Lemmink (University of Groningen). His presentation focused on
connecting sports sciences, data science and technology.
During lunch participants had the opportunity to a visit the wind tunnel facilities of
Eindhoven University of Technology, led by Professor Bert Blocken. In his key note
presentation ‘Marginal gains... and more’ Professor Blocken discussed the application of
aerodynamics in cycling. Participants probably were not aware at that time, but he also
gave a glimpse of the sub 2 hours marathon attempt that would take place one thought
later was given. The third keynote was given by Harry van Dorenmalen, Chairman of
the Dutch Top Team Sport. In his “Keep going, get better!” he went deeper into the
SportInnovator ecosystem and the importance of cooperation.
We would like to thank all participants and presenters for their enthusiasm and
contributions to crossing borders in research on sport and physical activity, as well to
Marly Sluijsmans for her support on organizing the conference together with all the
student volunteers involved.
Steven Vos, Juan Restrepo Villamizar, and Aarnout Brombacher
Editors
Koen Lemmink
Professor University Medical Center Groningen,
Universi-ty of Groningen.
Bert Blocken
Professor Eindhoven University of Technology &
KU Leuven
Harry van
Dorenmalen
Chairman TopTeam sport SportInnovator
Technology and data in
football: the missing link?
‘Marginal gains... and more’
Improving &
Optimizing
1 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation Extended abstract
Fully Automated Dynamic Ultrasound Skeletal Muscle
Analysis during Walking Exercise
Cristina Caresio1, Benjamin Tchang2, Jori Verbeek2, Frans van de Vosse1 and Richard Lopata1
1 Department of Biomedical Engineering, Eindhoven University of Technology, The Netherlands 2 USONO, Eindhoven, The Netherlands
Emails: C.Caresio@tue.nl – benjamin@usono.nl – jori@usono.nl – F.N.v.d.Vosse@tue.nl – R.Lopata@tue.nl
Submitted: 31/08/2019
Keywords: ultrasound imaging, voluntary contractions, walking exercise, automated algorithm, US probe
fixation
1
Introduction
Ultrasound (US) Imaging is a fundamental procedure in the assessment of skeletal muscle architecture. Dynamic contraction implies muscle anatomical deformation, which depends on exercise type and intensity. Muscle Thickness (MT) and Fascicles Pennation Angle (FPA) reflect the potential to generate force and have therefore been investigated in adaptations occurring with training, injuries and pathological conditions [1],[2]. Although previous works have examined the muscle architecture change during dynamic exercise, the information on deformation in realistic long-lasting non-isometric dynamic conditions are insufficient and incomplete [3], [4]. No continuous MT and FPA measurements has been proposed in literature, due to the lack of US probe fixation systems and the substantial effort required for manual measurements, which results to be time-consuming and prone of errors.
2
Research Question
In this study, we investigated MT and FPA measurements during walking using US. Thanks to the recent introduction of US probe fixation systems, skeletal muscles can be imaged with US in dynamic conditions [5],[6]. Furthermore, we propose an automated algorithm for US muscle architecture analysis in dynamic conditions, starting from previous developed automated algorithms [7], [8].
2 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
3
Material and Methods
8 healthy subjects (age: 24.5 ± 1.9 y, BMI: 22.8 ± 3.0 kg/m2) are asked to walk at 4 km/h on a treadmill (LifeFitness, Illinois, USA) as in Figure 1 - A. US videos of the medial gastrocnemius muscle (Figure 1 - C) are recorded at 20 Hz with a MyLab70 ultrasound device equipped with a linear LA523 transducer (Esaote, Maastricht, The Netherlands) and fixated on the calf using a Probefix Dynamic (USONO, Eindhoven, The Netherlands) (Figure 1 - B). Dynamic US images are analysed with an automated algorithm for the continuous MT and FPA measurements. Results are compared with manual measurements performed by an expert operator (Figure 2, Panel A - B).
4
Results
Experiments were successful in all subjects, providing US data and muscle parameters for sequences of 10 seconds. The percentage of incorrect automatic muscle segmentation is below 0.1%. The averaged MT is 15.4 ± 0.2 mm, ranging during gate between 9.6 - 21.5 mm. The averaged FPA is 13.1 ± 2.0ᵒ, ranging between 4.6 - 18.2ᵒ. Preliminary results of the manual validation show that the differences between the automatic and manual measurements are below 0.1 mm for MT and 2.5o for FPA respectively (Figure 2, Panel C). Automated analysis takes less than 0.8 second per image, compared to the 1.5 minutes of the manual annotation. The MT measurements presented a temporal pattern that can be qualitatively associated, for each subject, to a specific phase of the gait cycle (Figure 2, Panel D).
Figure 1. Example of
continuous ultrasound acquisition during walking exercise.
(A) Subject walking on the treadmill;
(B) Probefix Dynamic (USONO, Eindhoven); (C) Ultrasound image if the medial gastrocnemius acquired in dynamic condition.
2 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
3
Material and Methods
8 healthy subjects (age: 24.5 ± 1.9 y, BMI: 22.8 ± 3.0 kg/m2) are asked to walk at 4 km/h on a treadmill (LifeFitness, Illinois, USA) as in Figure 1 - A. US videos of the medial gastrocnemius muscle (Figure 1 - C) are recorded at 20 Hz with a MyLab70 ultrasound device equipped with a linear LA523 transducer (Esaote, Maastricht, The Netherlands) and fixated on the calf using a Probefix Dynamic (USONO, Eindhoven, The Netherlands) (Figure 1 - B). Dynamic US images are analysed with an automated algorithm for the continuous MT and FPA measurements. Results are compared with manual measurements performed by an expert operator (Figure 2, Panel A - B).
4
Results
Experiments were successful in all subjects, providing US data and muscle parameters for sequences of 10 seconds. The percentage of incorrect automatic muscle segmentation is below 0.1%. The averaged MT is 15.4 ± 0.2 mm, ranging during gate between 9.6 - 21.5 mm. The averaged FPA is 13.1 ± 2.0ᵒ, ranging between 4.6 - 18.2ᵒ. Preliminary results of the manual validation show that the differences between the automatic and manual measurements are below 0.1 mm for MT and 2.5o for FPA respectively (Figure 2, Panel C). Automated analysis takes less than 0.8 second per image, compared to the 1.5 minutes of the manual annotation. The MT measurements presented a temporal pattern that can be qualitatively associated, for each subject, to a specific phase of the gait cycle (Figure 2, Panel D).
Figure 1. Example of
continuous ultrasound acquisition during walking exercise.
(A) Subject walking on the treadmill;
(B) Probefix Dynamic (USONO, Eindhoven); (C) Ultrasound image if the medial gastrocnemius acquired in dynamic condition.
3 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
5
Conclusion
The proposed method shows that continuous automated US skeletal muscle architecture analysis during walking is feasible and has the potential of being robust and accurate, finding application in clinical practice and sports science.
Figure 2. Output of the automated
algorithm for skeletal muscle measurement.
(A) Example of manual muscle thickness measurements in five points and Pennation Angle along three representative fascicles of the Medial Gastrocnemius muscle. (B) Example of automatic Muscle Thickness and Pennation angle on the corresponding three Fascicles.
(C) Manual vs Automatic Muscle Thickness and Pennation Angle in 10 seconds of acquisitions. (D) Muscle Thickness changes associated with Gait Phases along 8 averaged cycles.
4 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
References
[1] M. V Narici, T. Binzoni, E. Hiltbrand, J. Fasel, F. Terrier, R. Fisiologia, T. Biomediche, and C. Nazionale, “In vivo human gastrocnemius architecture with changing joint angle at rest and during graded isometric contraction,” vol. 496, pp. 287–297, 1996.
[2] Y. Fukumoto, T. Ikezoe, Y. Yamada, R. Tsukagoshi, M. Nakamura, N. Mori, M. Kimura, and N. Ichihashi, “Skeletal muscle quality assessed from echo intensity is associated with muscle strength of middle-aged and elderly persons,” Eur. J. Appl. Physiol., vol. 112, pp. 1519–1525, 2012.
[3] L. Mademli and A. Arampatzis, “Behaviour of the human gastrocnemius muscle architecture during submaximal isometric fatigue,” Eur. J. Appl. Physiol., vol. 94, no. 5–6, pp. 611–617, 2005.
[4] S. Bohm, R. Marzilger, F. Mersmann, A. Santuz, and A. Arampatzis, “Operating length and velocity of human vastus lateralis muscle during walking and running,” Sci. Rep., vol. 8, no. 1, Dec. 2018.
[5] H. M. Heres, T. Schoots, B. C. Y. Tchang, M. C. M. Rutten, H. M. C. Kemps, F. N. van de Vosse, and R. G. P. Lopata, “Perfusion dynamics assessment with Power Doppler ultrasound in skeletal muscle during maximal and submaximal cycling exercise,” Eur. J. Appl. Physiol., vol. 118, no. 6, pp. 1209–1219, . 2018.
[6] Cristina Caresio, Benjamin Tchang, Jori Verbeek, Frans van de Vosse, and Richard Lopata, “Dynamic Ultrasound Skeletal Muscle Analysis,” Gait Posture, Suppl. Mater. ESMAC Conf. 2019, Run. best-paper Award.
[7] C. Caresio, M. Salvi, F. Molinari, K. M. Meiburger, and M. A. Minetto, “Fully Automated Muscle Ultrasound Analysis (MUSA): Robust and Accurate Muscle Thickness Measurement,” Ultrasound Med. Biol., vol. 43, no. 1, 2017.
[8] M. Salvi, C. Caresio, K. M. Meiburger, B. De Santi, F. Molinari, and M. A. Minetto, “Transverse Muscle Ultrasound Analysis (TRAMA): Robust and Accurate Segmentation of Muscle Cross-Sectional Area,” Ultrasound
Med. Biol., vol. 45, no. 3, pp. 672–683, 2019.
1 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation Extended abstract
The Running & Exercise Mental Break Optimisation
(REMBO) App: Design Process and User Evaluation.
Luuk P. van Iperen 1, Jan de Jonge 1,2,3, Josette M.P. Gevers 1, and Steven B. Vos 4,5,6
1 Human Performance Management Group, Eindhoven University of Technology, Eindhoven 5600 MB, The
Netherlands
2 Department of Social, Health, and Organisational Psychology, Utrecht University, Utrecht 3508 TC, The Netherlands 3 School of Psychology, Asia Pacific Centre for Work Health and Safety, University of South Australia, Adelaide SA
5001, Australia
4 Department of Industrial Design, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands 5 School of Sport Studies, Fontys University of Applied Sciences, 5644 HZ Eindhoven, The Netherlands 6 Policy in Sports and Physical Activity Research Group, University of Leuven, Leuven, Belgium
Emails: l.p.v.iperen@tue.nl; j.d.jonge@tue.nl; j.m.p.gevers@tue.nl; s.vos@tue.nl. Submitted: 11-09-2019
Keywords: running application; user experience evaluation; running-related injury; recovery; passion
1
Introduction
The popularity of running is steadily growing, which is great news considering its potential health benefits [1]. Less fortunate, however, is the associated injury rate in running. Recent research shows that no less than 52.2% of runners report a running-related injury (RRI) in the past 12 months [2]. These numbers suggests injury prevention in running is pivotal, but insufficiently practiced. The most common barriers for practicing injury prevention in running are “not knowing what to do” and “no history of RRI” [3] (p. 10). Therefore, we designed an application specifically for runners to prevent injuries and internalize a habitual preventative mindset concerning RRI in runners. We tested the application in a randomized controlled trial [4].
The application or app, which we aptly named the Running & Exercise Mental Break Optimisation (REMBO) app, contained various elements aimed at establishing the aforementioned goals, most importantly via a ‘running check’ questionnaire. The purpose of this short questionnaire was to determine ones’ personal training capacity via a set of questions and to give a personalized advice accordingly relating to their planned training load for that day. In doing so, its goal was to help runners stay within healthy boundaries of training, thereby preventing overtraining and the associated risk of injury. Thereby the app stimulated them to ensure adequate recovery had taken place before they engaged in running again. The main proposed workings of our app revolved around mental aspects of injury prevention, which are likely to be very important in injury prevention [4]. These aspects include physical, cognitive and emotional recovery [5], and obsessive and harmonious passion [6], which are described more in-depth in our design paper [4].
In order to evaluate the ease of use and effectiveness of this app we set out to gage user experiences. Using a semi-structured interview format, we set out to qualitatively explore how users experienced usage of the app and points for improvement they recommended. Note that predicted relations with injuries and/or results from the trial are not part of the current abstract and will be reported elsewhere.
2
Methods
As a summary of the app design and workings: we compared several online app designing platforms, eventually selecting one which facilitated such design via a (near-)drag & drop experience with basic HTML support. The earlier mentioned ‘running check’ consisted of 12 items mainly related to mental aspects, such as mental fatigue, feelings of obligation, and focus. Some items related to physical indicators were also included (e.g., joint pain). All items were rated on a 7-point Likert scale. The feedback mechanism on planned trainings based on this
4 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
References
[1] M. V Narici, T. Binzoni, E. Hiltbrand, J. Fasel, F. Terrier, R. Fisiologia, T. Biomediche, and C. Nazionale, “In vivo human gastrocnemius architecture with changing joint angle at rest and during graded isometric contraction,” vol. 496, pp. 287–297, 1996.
[2] Y. Fukumoto, T. Ikezoe, Y. Yamada, R. Tsukagoshi, M. Nakamura, N. Mori, M. Kimura, and N. Ichihashi, “Skeletal muscle quality assessed from echo intensity is associated with muscle strength of middle-aged and elderly persons,” Eur. J. Appl. Physiol., vol. 112, pp. 1519–1525, 2012.
[3] L. Mademli and A. Arampatzis, “Behaviour of the human gastrocnemius muscle architecture during submaximal isometric fatigue,” Eur. J. Appl. Physiol., vol. 94, no. 5–6, pp. 611–617, 2005.
[4] S. Bohm, R. Marzilger, F. Mersmann, A. Santuz, and A. Arampatzis, “Operating length and velocity of human vastus lateralis muscle during walking and running,” Sci. Rep., vol. 8, no. 1, Dec. 2018.
[5] H. M. Heres, T. Schoots, B. C. Y. Tchang, M. C. M. Rutten, H. M. C. Kemps, F. N. van de Vosse, and R. G. P. Lopata, “Perfusion dynamics assessment with Power Doppler ultrasound in skeletal muscle during maximal and submaximal cycling exercise,” Eur. J. Appl. Physiol., vol. 118, no. 6, pp. 1209–1219, . 2018.
[6] Cristina Caresio, Benjamin Tchang, Jori Verbeek, Frans van de Vosse, and Richard Lopata, “Dynamic Ultrasound Skeletal Muscle Analysis,” Gait Posture, Suppl. Mater. ESMAC Conf. 2019, Run. best-paper Award.
[7] C. Caresio, M. Salvi, F. Molinari, K. M. Meiburger, and M. A. Minetto, “Fully Automated Muscle Ultrasound Analysis (MUSA): Robust and Accurate Muscle Thickness Measurement,” Ultrasound Med. Biol., vol. 43, no. 1, 2017.
[8] M. Salvi, C. Caresio, K. M. Meiburger, B. De Santi, F. Molinari, and M. A. Minetto, “Transverse Muscle Ultrasound Analysis (TRAMA): Robust and Accurate Segmentation of Muscle Cross-Sectional Area,” Ultrasound
Med. Biol., vol. 45, no. 3, pp. 672–683, 2019.
1 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation Extended abstract
The Running & Exercise Mental Break Optimisation
(REMBO) App: Design Process and User Evaluation.
Luuk P. van Iperen 1, Jan de Jonge 1,2,3, Josette M.P. Gevers 1, and Steven B. Vos 4,5,6
1 Human Performance Management Group, Eindhoven University of Technology, Eindhoven 5600 MB, The
Netherlands
2 Department of Social, Health, and Organisational Psychology, Utrecht University, Utrecht 3508 TC, The Netherlands 3 School of Psychology, Asia Pacific Centre for Work Health and Safety, University of South Australia, Adelaide SA
5001, Australia
4 Department of Industrial Design, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands 5 School of Sport Studies, Fontys University of Applied Sciences, 5644 HZ Eindhoven, The Netherlands 6 Policy in Sports and Physical Activity Research Group, University of Leuven, Leuven, Belgium
Emails: l.p.v.iperen@tue.nl; j.d.jonge@tue.nl; j.m.p.gevers@tue.nl; s.vos@tue.nl. Submitted: 11-09-2019
Keywords: running application; user experience evaluation; running-related injury; recovery; passion
1
Introduction
The popularity of running is steadily growing, which is great news considering its potential health benefits [1]. Less fortunate, however, is the associated injury rate in running. Recent research shows that no less than 52.2% of runners report a running-related injury (RRI) in the past 12 months [2]. These numbers suggests injury prevention in running is pivotal, but insufficiently practiced. The most common barriers for practicing injury prevention in running are “not knowing what to do” and “no history of RRI” [3] (p. 10). Therefore, we designed an application specifically for runners to prevent injuries and internalize a habitual preventative mindset concerning RRI in runners. We tested the application in a randomized controlled trial [4].
The application or app, which we aptly named the Running & Exercise Mental Break Optimisation (REMBO) app, contained various elements aimed at establishing the aforementioned goals, most importantly via a ‘running check’ questionnaire. The purpose of this short questionnaire was to determine ones’ personal training capacity via a set of questions and to give a personalized advice accordingly relating to their planned training load for that day. In doing so, its goal was to help runners stay within healthy boundaries of training, thereby preventing overtraining and the associated risk of injury. Thereby the app stimulated them to ensure adequate recovery had taken place before they engaged in running again. The main proposed workings of our app revolved around mental aspects of injury prevention, which are likely to be very important in injury prevention [4]. These aspects include physical, cognitive and emotional recovery [5], and obsessive and harmonious passion [6], which are described more in-depth in our design paper [4].
In order to evaluate the ease of use and effectiveness of this app we set out to gage user experiences. Using a semi-structured interview format, we set out to qualitatively explore how users experienced usage of the app and points for improvement they recommended. Note that predicted relations with injuries and/or results from the trial are not part of the current abstract and will be reported elsewhere.
2
Methods
As a summary of the app design and workings: we compared several online app designing platforms, eventually selecting one which facilitated such design via a (near-)drag & drop experience with basic HTML support. The earlier mentioned ‘running check’ consisted of 12 items mainly related to mental aspects, such as mental fatigue, feelings of obligation, and focus. Some items related to physical indicators were also included (e.g., joint pain). All items were rated on a 7-point Likert scale. The feedback mechanism on planned trainings based on this
2 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
‘running check’ was implemented using traffic lights, a common and effective approach in interventions [7]; green for: safe running, orange for: risky running, and red for: no running recommended at all. Categories were determined via an algorithm based on ‘running check’ scores of users. Orange and red traffic lights were accompanied by advice on reduction of or alternatives for participants’ planned training.
The REMBO app was first tested in a randomized controlled trial [4]. After this trial we requested 37 people from the intervention group (n = 214) (i.e., those who had access to the app) to partake in an interview. Of those invited, 14 accepted and were interviewed in a semi-structured fashion by phone. During this interview 18 questions (i.e., closed, open, and follow-up questions based on certain answers) were used to explore the following facets of user experience: general perception of the app and the ‘run check’; outcomes resulting from app usage; anticipated future use; and possible improvements [8]. Results were analyzed according to Grounded Theory [9] using QDA Miner Lite (v2.0.6; Provalis Research, Montreal, Quebec, Canada).
3
Results
More positive than negative experiences were mentioned, with only a subset of these negative experiences pertaining to actual app content (c.f., app look). The ‘running check’ was nearly uniformly deemed a good indicator of their capacity, although some comments about its lack of physical questions and broadness of traffic light categories were mentioned.
A wide variety of ideas were offered when asking for improvements, including the ability to save data and link the app with other apps. Most interviewees said the app influenced their opinion of running injuries by increasing awareness of mental aspects (e.g., detaching from ones’ sport), followed by a smaller share which said the app had not changed anything, following by a variety of yet smaller shares mentioning various positive outcomes other than awareness. Participants were nearly uniform (86%) in saying that the app would not require recurring usage but that its mechanism was internalized after a period of usage during the trial.
4
Conclusion
The goal of this study was to qualitatively evaluate the REMBO app among its users. Generally, the app was received well and achieved some of its intended effects, such as increased awareness of mental aspects (e.g., mentally detaching from ones’ sport) of injury prevention. Multiple points for improvement were offered by users, including collaboration with or implementation in other apps and the option to save ones’ data.
As the goal of our app was rather small in scope (i.e., to test proposed mechanisms relating to RRI) our design process was not as elaborate as some similar studies [10]. Combining solely the functional mechanism of our application with other apps which already possesses adequate design and a benefitting user base may avoid issues pertaining to our basic design.
The qualitative nature of this study can be considered both a strong and weak point, as the exploratory nature allows us to explore aspects otherwise missed, but the very nature (and sample size) of such studies generally complicate generalizability. Furthermore, some of our findings can also be considered ambiguous due to contrasting desires with comparable amounts of proponents on both sides of some issues.
In conclusion, this study shows that the design and implementation of the REMBO app were received favorably among the interviewed runners and app usage resulted in increased awareness of the mental aspects (e.g., mental recovery, passion) of RRI prevention.
References
1. Shipway, R.; Holloway, I. (2010). Running free: Embracing a healthy lifestyle through distance running. Perspectives
in Public Health, 130, 270–276. doi:10.1177/1757913910379191
2. van Poppel, D.; Scholten-Peeters, G. G. M.; van Middelkoop, M.; Koes, B. W.; Verhagen, A. P. (2018). Risk models for lower extremity injuries among short- and long distance runners: A prospective cohort study. Musculoskeletal
Science and Practice, 36, 48–53. doi:10.1016/j.msksp.2018.04.007
3. Fokkema, T.; de Vos, R.-J.; Bierma-Zeinstra, S. M. A.; van Middelkoop, M. (2019). Opinions, Barriers, and Facilitators of Injury Prevention in Recreational Runners. Journal of Orthopaedic & Sports Physical Therapy, 1–22. doi:10.2519/jospt.2019.9029
3 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
4. de Jonge, J.; van Iperen, L.; Gevers, J.; Vos, S. (2018). “Take a Mental Break!” study: Role of mental aspects in running-related injuries using a randomised controlled trial. BMJ Open Sport & Exercise Medicine, 4, e000427. doi:10.1136/bmjsem-2018-000427
5. de Jonge, J.; Spoor, E.; Sonnentag, S.; Dormann, C.; van den Tooren, M. (2012). “Take a break?!” Off-job recovery, job demands, and job resources as predictors of health, active learning, and creativity. European Journal of Work and
Organizational Psychology, 21, 321–348. doi:10.1080/1359432x.2011.576009
6. Vallerand, R. J. (2010). On Passion for Life Activities. Advances in Experimental Social Psychology, 97–193. Elsevier. doi:10.1016/s0065-2601(10)42003-1
7. Thorgeirsson, T.; Kawachi, I. (2013). Behavioral Economics. American Journal of Preventive Medicine, 44, 185–189. doi:10.1016/j.amepre.2012.10.008
8. Vervuurt, B. (2019). Improving the REMBO app: Mental aspects of running-related injuries. Unpublished bachelor thesis, Eindhoven University of Technology, The Netherlands, 18-01-2019.
9. Bryman, A. (2012). Social research methods. Oxford New York: Oxford University Press.
10. Vos; S.; Janssen; M.; Goudsmit; J.; Lauwerijssen; C.; Brombacher; A. (2016). From Problem to Solution: Developing a Personalized Smartphone Application for Recreational Runners following a Three-step Design Approach. Procedia
3 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
4. de Jonge, J.; van Iperen, L.; Gevers, J.; Vos, S. (2018). “Take a Mental Break!” study: Role of mental aspects in running-related injuries using a randomised controlled trial. BMJ Open Sport & Exercise Medicine, 4, e000427. doi:10.1136/bmjsem-2018-000427
5. de Jonge, J.; Spoor, E.; Sonnentag, S.; Dormann, C.; van den Tooren, M. (2012). “Take a break?!” Off-job recovery, job demands, and job resources as predictors of health, active learning, and creativity. European Journal of Work and
Organizational Psychology, 21, 321–348. doi:10.1080/1359432x.2011.576009
6. Vallerand, R. J. (2010). On Passion for Life Activities. Advances in Experimental Social Psychology, 97–193. Elsevier. doi:10.1016/s0065-2601(10)42003-1
7. Thorgeirsson, T.; Kawachi, I. (2013). Behavioral Economics. American Journal of Preventive Medicine, 44, 185–189. doi:10.1016/j.amepre.2012.10.008
8. Vervuurt, B. (2019). Improving the REMBO app: Mental aspects of running-related injuries. Unpublished bachelor thesis, Eindhoven University of Technology, The Netherlands, 18-01-2019.
9. Bryman, A. (2012). Social research methods. Oxford New York: Oxford University Press.
10. Vos; S.; Janssen; M.; Goudsmit; J.; Lauwerijssen; C.; Brombacher; A. (2016). From Problem to Solution: Developing a Personalized Smartphone Application for Recreational Runners following a Three-step Design Approach. Procedia
Sports,
Data &
1 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation Extended abstract
COMPARING THE PERFORMANCE OF YOUTH
FOOTBALLERS ON ANTICIPATORY CAPABILITY
AND THEIR RESPECTIVE COACH RATINGS
Tom de Joode 1 and Saumya A. Mehta 1 and John van der Kamp 1 and Geert J. P. Savelsbergh 1
1 Research Institute Amsterdam Movement Science, Department of Human Movement Sciences, VU University, Amsterdam, The Netherlands
Emails: t.de.joode@vu.nl, saumyamehta8@gmail.com, j.vander.kamp@vu.nl, g.j.p.savelsbergh@vu.nl Submitted: 31-08-2019
Keywords: talent identification, decision-making, anticipation, actuarial judgment, perceptual-motor skill
1
Introduction
Recognizing athletes who will be successful in the future is the key of talent identification [1,2]. Selecting players is mainly done by scouts and coaches. While it is general accepted that they have an “eye” for talent, several studies show that their decision-making is affected by biases like individual experience [3] and relative age effect [4,5], even if coaches are made known of it [6]. Physiological and anthropometric characteristics of players cause unequal maturation within youth and cause performance differences between players [7]. Literature regarding talent identification suggest that perceptual skill is promising. Experts are found to be more efficient than novices in recognizing, analyzing and interpreting contextual information [8]. Moreover, more recent studies have emphasized how the perceptual capacity of experts focusses them more on important information of actions of opponents compared to novices. This higher anticipation capability lets experts gain better contextual information and act more efficiently by making more meaningful associations [9-11]. With the importance of anticipatory capability, the game-insight test (GIT) was developed and found to distinguish between skilled and less-skilled players [12]. This test might be an important tool to improve the talent identification. In current study the GIT is a tool for comparison between the coaches “eye” and the scorers for anticipation capability. Accordingly the GIT can be validated for clinical judgement and to gain knowledge whether a coach can differentiate players who can anticipate better or worse during a positional game situation. Thereby, it examines whether there is a correlation between coach ratings and the scores of players on the GIT.
2
Method
2.1 Participants
A total of 52 elite youth football players (mean age 11.6 years) participated in the study. Participants were playing in the squads of under 11, 12 and 13 of two major football academies at the highest level in the Netherlands. The questionnaire was answered by four head coaches of the respective teams.
2.2 Game-Insight Test
The GIT consisted of a 4 v 4 small sided game which were projected on a 2.4 x 2.4m screen. The videos were shot from the participant’s point of view. Every video consisted of a positional game as well as a pass at the end of the video. Based on previous research there were three types of videos [8-11]:
- The film stopped 80 milliseconds after the passer made contact with the ball. - The film stopped at the time when the passer made contact with the ball. - The film stopped 80 milliseconds before the passer made contact with the ball.
2 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
Furthermore, two types of actions leading up to the final pass were also applied in the videos: a “solo” situation where an individual had the ball and then made the final pass, and a “duel” situation of uncertainty regarding possession and the final passer of the ball. 30 such films were used for each participant. The score of a participant performance was obtained by: the number of times the participant moved in the right direction (left, right or center of the starting position), the right zone and the right time (compared to answer videos). 1 point was awarded to the participant for each of these parameters, results in possible maximum of 3 points per trial.
2.3 Questionnaire
The questionnaire consisted of a 5-point Likert scale. The coaches were asked to rate each player for Creativity, Decision-Making and Anticipation.
3
Results
With use of the means and standard deviation high (N=13) and low (N=11) groups were created based on the scores of the coaches. Rating and the Game-Insight Test were conducted for both the groups on scores over all 30 trials, as well as 21 trials (excluding fake and indistinguishable trials). No significant correlations were found when Coach Rating was compared to the Score on Game-Insight Test on 30 trials (r=-.059; p= .679), while a weak yet statistically significant correlation was found between the Coach Ratings and Scores on Game-Insight Test over 21 trials (r=.290; p=.037). The scores on 21 trails were used for further analysis. Table 1 shows that the high scoring group was associated with a higher mean score on the game-insight test as compared to the low scoring group.
Table 1. Mean scores on GIT over 21 trials as a function of high & low scores on coach rating (CR)
Group N Mean Std. Deviation
GIT High scorers CR 13 Low scorers CR 11 29.8462 25.1818 4.45058 2.85721
In addition, the high- and low scoring groups created by CR, were also analyzed for easier and more difficult trials. The results of the t-test show that the mean difference in scores of high and low scoring groups on Duel trials (M=7.31, SD=1 for high scorers and M=7.36. SD=2 for low scorers) is not significant (as shown in table 2). There is, however, a significant (t(22)=2.653, p=.015) difference in the solo trials as a function of the high and low scoring groups (M=18, SD= 4.2 for high scorers and M=13.8, SD=3.3 for low scorers).
Table 2. Solo and duel trials as function of high and low scoring groups
Group t df Sig. (2-tailed)
Duel -.085 22 .933 Solo 2.65 22 .015
4
Conclusion
The study aim was to examine whether there is a relationship between an objective measure of anticipation and the coaches judgement. It was assessed to serve as a method of further validating the game-insight test – the objective measure, while also exploring the possible applications of this to coaches for selecting players. The analysis revealed a significant difference in performance of high and low scorers according to coach ratings on the game-insight test. It also revealed that this difference held significance when easier “solo” trials were considered, whereas no significant difference was found on the harder “duel” challenges. Indeed, what the coach “sees” is validated by the results on the game-insight test. It might be difficult to grade anticipatory capability correctly for each individual at this age, but coaches can certainly distinguish the best and worst within their group. Therefore, combining and comparing the opinion of coaches with objective perceptual skill measure (GIT) might increase the productivity of talent identification program.
2 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
Furthermore, two types of actions leading up to the final pass were also applied in the videos: a “solo” situation where an individual had the ball and then made the final pass, and a “duel” situation of uncertainty regarding possession and the final passer of the ball. 30 such films were used for each participant. The score of a participant performance was obtained by: the number of times the participant moved in the right direction (left, right or center of the starting position), the right zone and the right time (compared to answer videos). 1 point was awarded to the participant for each of these parameters, results in possible maximum of 3 points per trial.
2.3 Questionnaire
The questionnaire consisted of a 5-point Likert scale. The coaches were asked to rate each player for Creativity, Decision-Making and Anticipation.
3
Results
With use of the means and standard deviation high (N=13) and low (N=11) groups were created based on the scores of the coaches. Rating and the Game-Insight Test were conducted for both the groups on scores over all 30 trials, as well as 21 trials (excluding fake and indistinguishable trials). No significant correlations were found when Coach Rating was compared to the Score on Game-Insight Test on 30 trials (r=-.059; p= .679), while a weak yet statistically significant correlation was found between the Coach Ratings and Scores on Game-Insight Test over 21 trials (r=.290; p=.037). The scores on 21 trails were used for further analysis. Table 1 shows that the high scoring group was associated with a higher mean score on the game-insight test as compared to the low scoring group.
Table 1. Mean scores on GIT over 21 trials as a function of high & low scores on coach rating (CR)
Group N Mean Std. Deviation
GIT High scorers CR 13 Low scorers CR 11 29.8462 25.1818 4.45058 2.85721
In addition, the high- and low scoring groups created by CR, were also analyzed for easier and more difficult trials. The results of the t-test show that the mean difference in scores of high and low scoring groups on Duel trials (M=7.31, SD=1 for high scorers and M=7.36. SD=2 for low scorers) is not significant (as shown in table 2). There is, however, a significant (t(22)=2.653, p=.015) difference in the solo trials as a function of the high and low scoring groups (M=18, SD= 4.2 for high scorers and M=13.8, SD=3.3 for low scorers).
Table 2. Solo and duel trials as function of high and low scoring groups
Group t df Sig. (2-tailed)
Duel -.085 22 .933 Solo 2.65 22 .015
4
Conclusion
The study aim was to examine whether there is a relationship between an objective measure of anticipation and the coaches judgement. It was assessed to serve as a method of further validating the game-insight test – the objective measure, while also exploring the possible applications of this to coaches for selecting players. The analysis revealed a significant difference in performance of high and low scorers according to coach ratings on the game-insight test. It also revealed that this difference held significance when easier “solo” trials were considered, whereas no significant difference was found on the harder “duel” challenges. Indeed, what the coach “sees” is validated by the results on the game-insight test. It might be difficult to grade anticipatory capability correctly for each individual at this age, but coaches can certainly distinguish the best and worst within their group. Therefore, combining and comparing the opinion of coaches with objective perceptual skill measure (GIT) might increase the productivity of talent identification program.
3 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
5
References
1. Reilly, T., Williams, A. M., Nevill, A. & Franks, A. A multidisciplinary approach to talent identification in soccer.
J. Sports Sci. (2000). doi:10.1080/02640410050120078
2. Williams, A. M. & Reilly, T. Talent identification and development in soccer. J. Sports Sci. (2000). doi:10.1080/02640410050120041
3. Christensen, M. K. ‘An eye for talent’: Talent identification and the ‘practical sense’ of top-level soccer coaches.
Sociol. Sport J. (2009). doi:10.1123/ssj.26.3.365
4. Mann, D. L., Dehghansai, N. & Baker, J. Searching for the elusive gift: advances in talent identification in sport.
Current Opinion in Psychology (2017). doi:10.1016/j.copsyc.2017.04.016
5. Helsen, W. F., Van Winckel, J. & Williams, A. M. The relative age effect in youth soccer across Europe. J. Sports
Sci. (2005). doi:10.1080/02640410400021310
6. Hill, B. & Sotiriadou, P. Coach decision-making and the relative age effect on talent selection in football. Eur.
Sport Manag. Q. (2016). doi:10.1080/16184742.2015.1131730
7. Vaeyens, R., Lenoir, M., Williams, A. M. & Philippaerts, R. M. Talent identification and development programmes in sport: Current models and future directions. Sports Medicine (2008). doi:10.2165/00007256-200838090-00001 8. Abernethy, B. & Russell, D. G. Expert-Novice Differences in an Applied Selective Attention Task. J. Sport
Psychol. (1987). doi:10.1123/jsp.9.4.326
9. Dicks, M., Davids, K. & Button, C. Individual differences in the visual control of intercepting a penalty kick in association football. Hum. Mov. Sci. (2010). doi:10.1016/j.humov.2010.02.008
10. Savelsbergh, G. J. P., Van der Kamp, J., Williams, A. M. & Ward, P. Anticipation and visual search behaviour in expert soccer goalkeepers. in Ergonomics (2005). doi:10.1080/00140130500101346
11. Savelsbergh, G. J. P., Onrust, M., Rouwenhorst, A. & Van Der Kamp, J. Visual search and locomotion behaviour in a four-to-four football tactical position game. Int. J. Sport Psychol. (2006).
12. Savelsbergh, G. J. P., Haans, S. H. A., Kooijman, M. K. & van Kampen, P. M. A method to identify talent: Visual search and locomotion behavior in young football players. Hum. Mov. Sci. 29, 764–776 (2010).
1 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation Extended abstract
Inspirun app: User Test of an Algorithm that
Automatically adjusts Training Sessions for Runners
Mark Janssen 1,2, Jos Goudsmit 1,2 and Steven Vos 1,2
1 Eindhoven University of Technology, Industrial Design, Eindhoven, 5600, The Netherlands 2 Fontys University of Applied Science. Eindhoven, 5600, The Netherlands
Emails: m.a.janssen@tue.nl, j.f.a.goudsmit@tue.nl, s.vos@tue.nl Submitted: 01-09-2019
Keywords: Running App, Personalization, Motivation, User tests, Alogrithm
1
Introduction
Research shows that approximately 50-75% of (event) runners use a running app, especially novice or less experienced runners [7]. This trend of using apps while exercising is consistent with trends like Quantified Self [9] and mHealth [5], which emphasize the potential of these monitoring devices to contribute to an active lifestyle by supporting behavior change [6]. Unfortunately, among runners there is still a high drop-out rate due to injuries [1,2] and decrease in motivation. The majority of these runners lack personalized guidance and support. (e-)Coaching and tailored advice could help these runners to avoid injuries, set goals and maintain good intentions [11]. In particular, apps have several advantages. Smartphones are widely used, they are embedded in daily life [3,11], and allow people to collect data anywhere and anytime [4]. Since most people already have a smartphone (up to 76% of all adults) [10] and apps are relatively cheap, apps became accessible for almost everyone. Unfortunately, available running apps primarily focus on monitoring performance and offer few or no features that support tailored guidance (e.g. personalized training schemes) [8].
Therefore, Vos et al. designed Inspirun, an e-coach app for runners [11]. Inspirun is an intelligent running app that provides automatic adjustment of training schemes based on heart rate data and GPS-data. (see Vos et al. [11] for a full description of the design). The present paper aims to give insight into how end-users experience the personalization of the algorithm that automatically adjusts the training scheme. We expected that if the end-user follows the app's instructions and completes the training as prescribed, the training will be consistent with the personal load capacity of the user and will be perceived by the user as personalized.
2 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
2
Study protocol
After ethical approval, 43 runners participated in our study between spring 2018 and autumn 2018. The participants used Inspirun until they completed 20 training sessions (i.e. scheme). We used online questionnaires (every two weeks), and data collected by the Inspirun to monitor the participants over time.
A total of 36 participants logged one or more training session those who did not register any run, stated that they underestimated the time investment needed. After the second questionnaire (3rd week), 28 participants
remained. The main reason to quit was a bug that randomly caused disconnection with the Bluetooth heart rate sensor. These participants did not like the struggle of reconnecting with Bluetooth heart rate sensor while running.
More runners dropped out later in the study. Reasons mentioned for quitting were, being on holiday and hot temperatures since the study partly was conducted in the summer. Finally, 20 participants performed between 9 and 25 training sessions in the testing period. There is quite a variety in the number of questionnaires filled in, due to differences in running frequency per week.
3 Results
In the analysis, we included the results of the 20 participants who completed the full testing period. On average (including all questionnaires (n=95)) the experienced personalization scored a 3.81 (SD 0.76) on a 5-point Likert scale (see Table 1). Subsequently, we asked the runners to explain their scores. Those who were positive about the personalization (scoring 4 or 5) experienced the gradual increase of the intensity as pleasant. They felt that the sessions were challenging without being too hard. Runners who scored 2 or 3 (no 1 scores were given), indicated that they were not quite sure if the app accurately measured their heart rate or speed. Therefore, they doubted the accuracy of the personalization. We checked these claims by analyzing the data logged by Inspirun and indeed, for all runners who scored 2 or 3, on multiple runs pieces of the heart rate or GPS data were missing.
Table 1. Experience of the personalization of the training session on a 5-point Likert scale Two-weekly questionnaire
1 2 3 4 5 6 7 8
Mean score 3,85 3,65 3,90 4,11 3,63 3,17 4,00 4,00
SD 0,65 0,79 0,77 0,66 0,70 0,90 0,00 0,00
Number of runners 20 20 20 18 8 6 2 1
4
Discussion and conclusion
Inspirun was designed [11] to provide personalized training schemes based on biofeedback and GPS-data. The primary conclusion of our study is that the algorithm which automatically adapts training sessions to the runners’ physical load (if provided with accurate data) is experienced as personalized.
3 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
Participants, whose data was complete and accurate, stated that the sessions matched with their training level. Building a PRP on both subjective (RPE) as objective (HR and Speed) aspects of intensity seems to be a good combination to develop an algorithm that is sensible for change.
Runners whose data was incomplete or inaccurate, doubted whether the sessions were accurate. In particular, gaps in heart rate data caused problems in the automatic calculation of the next session. For example, some runners had problems with the Bluetooth connection between the heart rate monitor and their smartphone while running. Therefore, the app not always collected heart rate data, but instead wrote zeros in the dataset. This caused a much lower average heart rate due to the inclusion of the zeros. Consequently, making the app think that the prescribed heart rates were too high. The same goes for inaccuracies in running speed. Runners experienced some problems when running in wooded and hilly areas, resulting in lower average speeds than expected, making the app think that the session was too hard to complete when this was actually due to the environment or signal strength of the GPS.
The algorithm of creating automatically and personalized training sessions seems to work when provided with complete and accurate data. Although the robustness of the algorithm, especially how it deals with flaws in the dataset is not reliable yet. In future work, first of all, the Bluetooth connection must be improved. Second, to make the algorithm more robust, the algorithm must be adjusted so that incomplete datasets (e.g. with too many zeros) are not used in the PRP. Third, additional parameters like elevation and running surface must be included when calculating the speed of a session.
5 References
[1] Steven W Bredeweg. 2014. Running related injuries The effect of preconditioning program and
biomedical risk factors (Doctoral Dissertation). University of Groningen, Groningen, The Netherlands,
02-04-2014.
[2] Ida Buist, Steven W Bredeweg, Bram Bessem, Willem van Mechelen, Koen A P M Lemmink, and Ron L Diercks. 2010. Incidence and risk factors of running-related injuries during preparation for a 4-mile recreational running event. Br. J. Sports Med. 44, 8 (2010), 598–604. DOI:https://doi.org/10.1136/bjsm.2007.044677
[3] Joan Dallinga, Mark Janssen, Jet Van Der Werf, Ruben Walravens, Steven Vos, and Marije Deutekom. 2018. Analysis of the features important for the effectiveness of physical activity–related apps for recreational sports: Expert panel approach. J. Med. Internet Res. 20, 6 (2018). DOI:https://doi.org/10.2196/mhealth.9459
[4] Xavier Ferre, Elena Villalba, Hector Julio, and Hongming Zhu. 2017. Extending Mobile App Analytics for Usability Test Logging. In Human-Computer Interaction INTERACT, 114–131. DOI:https://doi.org/10.1007/978-0-387-35175-9
[5] Maddalena Fiordelli, Nicola Diviani, and Peter J Schulz. 2013. Mapping mHealth research: a decade of evolution. J. Med. Internet Res. 15, 5 (May 2013), e95. DOI:https://doi.org/10.2196/jmir.2430 [6] Liam G Glynn, Patrick S Hayes, Monica Casey, Fergus Glynn, Alberto Alvarez-Iglesias, John Newell,
Gearóid OLaighin, David Heaney, Martin O’Donnell, and Andrew W Murphy. 2014. Effectiveness of a smartphone application to promote physical activity in primary care: the SMART MOVE randomised
4 of 4
SECSI 2019 – Science & Engineering Conference on Sports Innovation
controlled trial. Br. J. Gen. Pract. 64, 624 (July 2014), e384-91. DOI:https://doi.org/10.3399/bjgp14X680461
[7] Mark Janssen, Jeroen Scheerder, Erik Thibaut, Aarnout Brombacher, and Steven Vos. 2017. Who uses running apps and sports watches? Determinants and consumer profiles of event runners’ usage of running-related smartphone applications and sports watches. PLoS One 12, 7 (July 2017), e0181167. DOI:https://doi.org/10.1371/journal.pone.0181167
[8] Anouk Middelweerd, Julia S Mollee, Nathalie C van der Wal, Johannes Brug, and Saskia J te Velde. 2014. Apps to promote physical activity among adults: a review and content analysis. Int. J. Behav.
Nutr. Phys. Act. 11, 97 (2014). DOI:https://doi.org/10.1186/s12966-014-0097-9
[9] Melanie Swan. 2012. Sensor Mania! The Internet of Things, Wearable Computing, Objective Metrics, and the Quantified Self 2.0. J. Sens. Actuator Networks 1, (2012), 217–253. DOI:https://doi.org/10.3390/jsan1030217
[10] Kyle Taylor and Laura Silver. 2019. Smartphone Ownership is grwoing rapidly around the world, but
not always equally. Washington DC. Retrieved from
https://www.pewglobal.org/wp-content/uploads/sites/2/2019/02/Pew-Research-Center_Global-Technology-Use-2018_2019-02-05.pdf [11] Steven Vos, Mark Janssen, Jos Goudsmit, Coen Lauwerijssen, and Aarnout Brombacher. 2016. From problem to solution: A three-step approach to design a personalized smartphone application for recreational runners. Procedia Eng. ISEA 00, (2016), 1–7.
1 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation Extended abstract
Automatic extraction of performance metrics from
football players with data mining
Josá Carlos Coutinho 1 and Marck de Greeff 2 and Cláudio Rebelo de Sá 3 and Nicolette Schipper-van
Veldhoven 2
1 Leiden University, Leiden, Netherlands
2 Windesheim University of Applied Sciences, Zwolle, Netherlands 3 University of Twente, Enschede, Netherlands
Emails: j.c.milheirosoarescoutinho@utwente.nl, j.w.de.greeff@windesheim.nl, c.f.pinhorebelodesa@utwente.nl,
n.schippervanveldhoven@windesheim.nl
Submitted: 09 September 2019
Keywords: football; data mining; unsupervised; kids; sports
1
Introduction
Organized youth sport is seen as an important socializing context for children and adolescents. Sports, which are also known as the third pedagogical environment (next to home and school), contribute to a range of positive outcomes: self-esteem, social behavior and integration. There is recent evidence that there is a positive relationship between ‘club culture and atmosphere’ and ‘organizational performance’ (Schoot, 2016). This means that when a sport club focuses more on the elements of club culture, financial performance, member satisfaction and social sport performance can increase.
Based on the literature we can distinguish several conditions for a pedagogical sport climate (PSC) (Schipper-van Veldhoven, 2016). For example, the focus on sport pleasure, as it is the number one motivation to participate in sport and therefore an important determining factor of (continuous) sport participation. Fun is also an important educational tool. In other words, a bad pedagogical climate can interfere with the performance in the practice of sports. This can be assessed that by measuring how they are performing in comparison with their peers.
Typically, in high performance sports, the positional data comes with event data, which can support more detailed analysis of the performance of player. However, labelled football (soccer) data is hard to acquire and it usually needs humans to annotate the match events. This process makes it more expensive to be obtained by smaller clubs.
For this reason, we developed an Unsupervised Football Analytics Tool – UnFOOT. UnFOOT combines data mining techniques and basic statistics to measure the performance of players and teams only from positional data. The capabilities of the tool involve preprocessing the match data, extraction of features, visualization of player and team performance. It also has built-in data mining techniques, such as association rule mining, subgroup discovery and a proposed approach to look for frequent distributions.
Due to the lack of data from children playing, we tested the proposed approach in data from six professional football matches.
2
UnFOOT
There already exist tools that given the positional and event-labeled data can extract useful knowledge from the teams and players (Bialkowskiet al., Gud-mundssonet al.). However, these tools require event-labeled data, which can be more expensive to obtain than positional data.UnFOOT uses positional data from players in a football match and extracts different statistics as well as performance indicators of players and teams.
2 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
UnFOOT offers a simple and intuitive GUI for analyzing football matches only from spatiotemporal data of players. (A demonstration can be watched at https://youtu.be/x86tg48qEs4). The pipeline involves 3 stages: Processing of the data, Representation and Data Mining.
After loading the data, the tool makes one pass on the data and outputs a new dataset with extracted features. These features include the distance covered, the speed and the acceleration of the players. The dataset is divided into time windows of the same size. For each window, several internal modules extract different performance indicators and statistics from the positional data. One of the metrics, pressure uses a clustering technique (DBSCAN), from the python package scikit-learn (https://scikit-learn.org/stable/modules/clustering.html). With the clusters, we are able to identify moments of higher pressure of the players during the match. In the end of the analysis, the overall and detailed results are stored into a .csv file to enable further analysis outside of the tool.
From these performance indicators UnFOOT produces an overall player score which is the mean of the indicators. These player scores are also added together to obtain the score of each team.
The UnFOOT tool has an interface with several data mining techniques to explore the features extracted. One module uses association rules mining to find relationships of performance indicators between consecutive time windows for a selected player. The last method uses subgroup discovery to find subgroups with unusual behaviour relatively to an user defined target. For the association rules mining module, we used
mlxtend (http://rasbt.github.io/mlxtend/), and for the subgroup discovery module we used pysubgroup (https://pypi.org/project/pysubgroup/).
Besides, we also use a method to look for frequent distributions. These distributions can represent speed or distance covered by players. It is similar to frequent pattern mining, except that the items are distributions. For that, we use the Kolmogorov-Smirnov (KS) to verify if the distribution of one player is significantly different from the other players. Then, UnFOOT counts how many times each distinct distribution is observed during the match to obtain the support (frequency) per player.
2.1 Representation and Structure
The GUI is divided in 4 different tabs: Player, Team, Data Analysis and Settings. Player, evaluates and compares players according to their overall score or specific performance indicators. (Figure 1 a)); Team, displays and compares different team scores and shows the best players (Figure 1 b)); Data Analysis, allows the user to use an interface to execute data mining algorithms on the match data; and Settings, lets the user load the dataset and define some basic settings before starting the analysis.
3
Results
Six professional football games were analyzed with the tool. Due to privacy issues, we are not able to provide more details about the match, such as the name of the best player per match or the names of the teams. According to some metrics obtained (Table 1), the best players of the match are usually found on the tool's top three players of the winning team. In two cases, they even had the best score overall. Even though the overall score was not originally designed to predict the best player of the match, we use it to validate the scoring
2 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
UnFOOT offers a simple and intuitive GUI for analyzing football matches only from spatiotemporal data of players. (A demonstration can be watched at https://youtu.be/x86tg48qEs4). The pipeline involves 3 stages: Processing of the data, Representation and Data Mining.
After loading the data, the tool makes one pass on the data and outputs a new dataset with extracted features. These features include the distance covered, the speed and the acceleration of the players. The dataset is divided into time windows of the same size. For each window, several internal modules extract different performance indicators and statistics from the positional data. One of the metrics, pressure uses a clustering technique (DBSCAN), from the python package scikit-learn (https://scikit-learn.org/stable/modules/clustering.html). With the clusters, we are able to identify moments of higher pressure of the players during the match. In the end of the analysis, the overall and detailed results are stored into a .csv file to enable further analysis outside of the tool.
From these performance indicators UnFOOT produces an overall player score which is the mean of the indicators. These player scores are also added together to obtain the score of each team.
The UnFOOT tool has an interface with several data mining techniques to explore the features extracted. One module uses association rules mining to find relationships of performance indicators between consecutive time windows for a selected player. The last method uses subgroup discovery to find subgroups with unusual behaviour relatively to an user defined target. For the association rules mining module, we used
mlxtend (http://rasbt.github.io/mlxtend/), and for the subgroup discovery module we used pysubgroup (https://pypi.org/project/pysubgroup/).
Besides, we also use a method to look for frequent distributions. These distributions can represent speed or distance covered by players. It is similar to frequent pattern mining, except that the items are distributions. For that, we use the Kolmogorov-Smirnov (KS) to verify if the distribution of one player is significantly different from the other players. Then, UnFOOT counts how many times each distinct distribution is observed during the match to obtain the support (frequency) per player.
2.1 Representation and Structure
The GUI is divided in 4 different tabs: Player, Team, Data Analysis and Settings. Player, evaluates and compares players according to their overall score or specific performance indicators. (Figure 1 a)); Team, displays and compares different team scores and shows the best players (Figure 1 b)); Data Analysis, allows the user to use an interface to execute data mining algorithms on the match data; and Settings, lets the user load the dataset and define some basic settings before starting the analysis.
3
Results
Six professional football games were analyzed with the tool. Due to privacy issues, we are not able to provide more details about the match, such as the name of the best player per match or the names of the teams. According to some metrics obtained (Table 1), the best players of the match are usually found on the tool's top three players of the winning team. In two cases, they even had the best score overall. Even though the overall score was not originally designed to predict the best player of the match, we use it to validate the scoring
3 of 3
SECSI 2019 – Science & Engineering Conference on Sports Innovation
function. However, this scoring function can only reasonably assess the quality of players, which are not goalkeepers. This is seen in Game 5, where the best player was actually a goalkeeper.
We can also observe that the sum of the team players' individual performance may not be enough to evaluate the performance of the team, since in only half of the cases the highest team score corresponds to the winning team.
Table 1. Results obtained with UnFOOT.
Match Winner Team A score Team B score Rank of the best player
1 A 758 778 3rd of Team A 2 A 814 811 1st overall 3 A 795 805 3rd of Team A 4 B 832 855 3rd of Team B 5 A 813 796 Last overall 6 A 816 819 1st overall (a) (b)
Figure 1. This figure shows an example of the interface of UnFOOT for the (a) Individual Players; (b) Teams
Acknowledgements
This work was financed by the project Kids First, project number 68639
References
1. Schoot, T.C. (2016). More than facilitating football. A study on the relationship between organizational practices and organizational performance of Dutch amateur football clubs. Tilburg: Tilburg University.
2. Schipper-van Veldhoven, N. (2016). Sport en lichamelijke opvoeding in pedagogisch perspectief, een gouden kans (Lectorale rede). Zwolle: Windesheimreeks kennis en onderzoek nr. 60
3. Bialkowski, A., Lucey, P., Carr, P., Yue, Y., Sridharan, S., Matthews, I.: Identifying team style in soccer using formations learned from spatiotemporal tracking data. In: 2014 IEEE International Conference on Data Mining Workshop. pp. 9–14 (Dec2014). https://doi.org/10.1109/ICDMW.2014.167
4. Gudmundsson, J., Wolle, T.: Football analysis using spatio-temporal tools. Computers, Environment and Urban Systems47, 16 – 27 (2014). https://doi.org/10.1016/j.compenvurbsys.2013.09.004, progress inMovement Analysis Experiences with Real Data
